Research on Intelligent Algorithms of Deep Learning in English Translation Quality Assessment

Since the unsupervised quality assessment is mostly measured by the inherent knowledge of machine translation, that is, through training the final quantitative results, it can be added to the evaluation process as a feature, so as to effectively solve the problem of incompleteness caused by different purposes. Quality assessment refers to the objective evaluation of the quality of the translated text when there is no reference text, which has high requirements for machine translation. This project intends to solve problems such as insufficient labeled learning samples, inconsistent learning objectives at two levels, unclear task objectives and incomplete feature learning in unsupervised quality assessment, etc. It intends to introduce parameters in unsupervised quality assessment into supervised quality assessment model as features to complement each other. The high and small resource comparison experiment of WMT2020 verified that the combination of the two models can better improve the accuracy of translation quality assessment, and the correlation with Pearson is also improved compared with the guided and unguided models under baseline. It is proved that the proposed algorithm can effectively improve the relationship between machine scoring and manual scoring in Chinese translation proficiency evaluation.